[cfe-dev] Installing libc++

[Howard Hinnant]

On Apr 24, 2011, at 1:11 PM, Jon Kalb wrote:

> On 4/24/11 9:01 AM, "Howard Hinnant" <hhinnant at apple.com> wrote:
> 
>> On Apr 24, 2011, at 11:51 AM, Jon Kalb wrote:
>> 
>>> 
>>> Thanks Howard, you guessed right.
>>> 
>>> This is my testit result (summary):
>>> 
>>> ****************************************************
>>> Results for /Users/jonkalb/Documents/development/developement/libcxx/test:
>>> using Apple clang version 2.0 (tags/Apple/clang-138) (based on LLVM 2.9svn)
>>> Target: x86_64-apple-darwin10
>>> Thread model: posix
>>> with -std=c++0x -stdlib=libc++
>>> ----------------------------------------------------
>>> sections without tests   : 1
>>> sections with failures   : 153
>>> sections without failures: 910
>>>                      +   ----
>>> total number of sections : 1064
>>> ----------------------------------------------------
>>> number of tests failed   : 1221
>>> number of tests passed   : 3019
>>>                      +   ----
>>> total number of tests    : 4240
>>> ****************************************************
>>> 
>>> Is this what I should expect?
>> 
>> I would've expected significantly fewer test failures for clang -std=c++0x
>> -stdlib=libc++:
>> 
>> ****************************************************
>> Results for /test:
>> using Apple clang version 2.0 (trunk 125236) (based on LLVM 2.9svn)
>> Target: x86_64-apple-darwin10
>> Thread model: posix
>> with -std=c++0x -stdlib=libc++ -arch x86_64 -U__STRICT_ANSI__ -I../include
>> /usr/lib/libc++.dylib
>> ----------------------------------------------------
>> sections without tests   : 1
>> sections with failures   : 74
>> sections without failures: 989
>>                       +   ----
>> total number of sections : 1064
>> ----------------------------------------------------
>> number of tests failed   : 215
>> number of tests passed   : 4023
>>                       +   ----
>> total number of tests    : 4238
>> ****************************************************
>> 
>> However it has been a couple of months since I actually ran the full test
>> suite (been busy elsewhere).  Did you get libc++abi.dylib from here:
>> 
>> http://home.roadrunner.com/~hinnant/libcppabi.zip
>> 
>> ?
>> 
>> http://libcxx.llvm.org/ says:
>> 
>>> To build on Mac OS X 10.6, you need a helper library and header found here.
>>> cp cxxabi.h to /usr/include, and cp libc++abi.dylib to /usr/lib.
>> 
>> 
>> This library implements things like exception handling.  Lack of this library
>> might explain your failure rate.
>> 
>> If that isn't it, email me privately your entire results and I'll take a look.
>> 
>> Howard
> 
> I noticed this difference in our reports:
> 
>> using Apple clang version 2.0 (tags/Apple/clang-138) (based on LLVM 2.9svn)
> using Apple clang version 2.0 (trunk 125236) (based on LLVM 2.9svn)
> 
> I think I was getting the clang++ in /usr/lib/ instead of in
> /Users/jonkalb/Documents/development/developement/build/Debug+Asserts/bin/
> 
> Could that be the issue?

Could be, I was wondering the same thing.  I get my clang from an internal daily build instead of from a release.

> I've changed my PATH so I now get:
> 
> Jon-Kalbs-MacBook-Pro:test jonkalb$ which clang++
> /Users/jonkalb/Documents/development/developement/build/Debug+Asserts/bin/cl
> ang++
> 
> and I'm running testit again. I will send the entire result to you off-list
> if it isn't closer to your result.

If it is the difference in clang's, turning off -std=c++0x could actually help as far as passing tests goes.  The test suite is looking for that flag and tries to test more stuff when it sees it.

Howard